Agentic commerce isn’t a 2030 problem. It’s a 2026 problem with a 2015 backend.
Everyone keeps pointing to the Morgan Stanley prediction. Half of online shoppers using AI agents by 2030. Cool timeline. Very comforting.
Meanwhile the agents are already here. And most commerce backends are answering the door in a bathrobe with no idea what day it is.
I’ve spent over a decade building commerce platforms. I’ve seen every wave ... mobile, headless, composable, MACH. Every time, the industry had a few years to prepare. This time it doesn’t.
What Happens When Teams Try to Add Agent Support
Here’s the pattern. I’ve watched it play out for 15 years, and it’s happening right now with agentic commerce.
A product team sees the headlines. AI traffic to retail sites up 805% on Black Friday. ChatGPT driving 20% of referral traffic to Walmart. They call a meeting. “We need agent support.” Engineering nods. The protocol documentation looks clean. “This is just an API layer.”
It’s never just an API layer.
Month 1. Engineering starts building the integration. Spirits are high.
Month 2. The team discovers the product feed is 6 hours stale. Agents need real-time. What was “good enough” for a human browsing experience is completely inadequate for machine-speed commerce. Your product catalog was designed for browsing, not querying. An agent asks for a waterproof hiking boot under $200 in size 11. Your search returns snow boots and sandals because your data was tagged for humans who can squint and figure out what “BLK RUNR SHO M 10.5” means. A human sees that and thinks “black running shoe, men’s, 10.5.” An agent sees gibberish and moves on.
Month 3. The checkout flow assumes a human is filling out forms. The entire purchase path was built around UI interactions. An agent needs to programmatically create, price, and complete a purchase without rendering a single page. Your promotions engine throws an error nobody ever tested against a non-human session.
Month 4. Inventory can’t answer availability questions programmatically. Real-time means real-time. Not “we update every 15 minutes.” An agent that checks availability and gets a false positive doesn’t shrug and try again. It tells the customer the item isn’t available and recommends a competitor. No second chance. No angry email. Just gone.
Month 5. “We need to rebuild our entire commerce API layer.”
Month 6. “This is going to take 18 months.”
Every time. New protocol launches. Teams think it’s an integration sprint. Turns out it requires infrastructure redesign. Timeline explodes. Competitors who started earlier compound their lead.
And here’s the part that makes it worse ... the agent doesn’t behave like a frustrated human customer. It doesn’t complain. Doesn’t leave a review. Doesn’t call customer service. It just never recommends you again. With humans, you get second chances. Agents optimize. And optimizing means routing around you permanently.
The Problem Isn’t the Protocol. It’s the Plumbing.
Google launched Universal Commerce Protocol at NRF with Walmart, Shopify, Target, and Wayfair. OpenAI shipped Agentic Commerce Protocol with Stripe. Amazon blocked both and built their own closed system.
Three billion-dollar bets on who owns the front door of agentic commerce. And the whole time I’m watching this unfold thinking … you guys know the front door isn’t the hard part, right?
This is like arguing over who makes the best doorbell while the house has no plumbing.
Protocols don’t fulfill orders. Protocols don’t calculate tax across 40 countries. Protocols don’t handle the return logic your ops team spent six months building because customers in Germany have different refund rights than customers in Texas.
I’ve built commerce platforms where we couldn’t even get order data to our OMS and downstream systems without hitting our heads against the wall because we didn’t have budget for a proper event bus. Duct tape and creative problem solving holding it together because the infrastructure investment was never prioritized.
Now we’re going to serve autonomous agents that need real-time everything across every system in the stack?
That’s like inviting Gordon Ramsay to your restaurant when you’re cooking on a hot plate and storing food in a cooler you borrowed from your neighbor.
The brands that win agentic commerce won’t be the ones who adopted the right protocol first. They’ll be the ones who fixed the boring stuff nobody wants to talk about at conferences.
What Actually Needs to Be True
Forget protocols for a minute. Before you can support any agent ... UCP, ACP, or whatever ships next quarter ... your platform needs four things. And none of them are glamorous.
Structured product data that machines can parse. Not marketing copy. Not abbreviated SKU descriptions. Precise, normalized attributes with consistent taxonomy across your entire catalog. Size, color, material, compatibility, use case ... all machine-readable. If your product data wasn’t built for machines, it doesn’t matter which protocol you implement. The agent will ask a question and your catalog will give a useless answer.
Real-time inventory accuracy. Not “in stock” badges that update hourly. Actual, second-by-second inventory positions. The difference between “accurate within an hour” and “accurate now” is the difference between a completed transaction and a customer permanently routed to your competitor.
Programmatic pricing and promotions. Your promotions engine can’t assume a human session. An agent needs to apply a coupon, calculate a bundle discount, and return a final price through an API call. If your promotions logic lives in front-end JavaScript, you have a problem that no protocol integration will solve.
Fulfillment orchestration that returns instant answers. Shipping options, delivery estimates, return policies ... all queryable, all real-time, all API-first. An agent asking “when will this arrive” needs an answer in milliseconds, not a page render.
These aren’t protocol requirements. They’re infrastructure requirements. And they’re the same requirements whether you go with UCP, ACP, both, or something that doesn’t exist yet.
Your platform either has this infrastructure or it doesn’t. If it doesn’t, both protocols are impossible. If it does, both protocols are straightforward integrations.
The protocol debate is a distraction. The infrastructure is the work.
The Nuance Nobody’s Talking About
Here’s where I think the entire agentic commerce conversation goes wrong.
A principal architect I was talking to recently reminded me of something an old mentor used to say ... “Skaters are gonna skate. Surfers are gonna surf.”
We’re building agentic commerce as though every customer is a skater. Fast. Efficient. Get the answer. Complete the transaction. And for a percentage of customers, that’s exactly right.
Skaters want speed. They know what they need. They want the fastest path to checkout. AI agents are perfect for them. “Buy the same detergent I bought last month.” Done. Replenishment buying. Known products. Specification-driven purchases where speed matters and exploration doesn’t.
Surfers want discovery. They’re browsing. Discovering. Building intent as they shop. AI agents rob them of that process. “Show me summer dresses” isn’t the same as scrolling through options, seeing something unexpected, and building purchase intent through exploration. If you force surfers through an agent, you lose them. Not because the agent failed. Because the agent succeeded at the wrong job.
This matters because it changes the architecture conversation entirely.
For 20 years, commerce platforms competed on the storefront. Best UX. Fastest page loads. Beautiful product pages. Agents don’t see any of that. They query your backend. Your storefront is invisible to them.
But surfers still need storefronts. They still need the browsing experience. They still need the product pages and the navigation and the discovery that builds purchase intent.
Which means your platform doesn’t just need to support agents. It needs to support agents AND traditional experiences AND the intelligence to determine which one a given customer needs at a given moment. The skaters get agents. The surfers get storefronts. And your infrastructure needs to serve both without treating either as an afterthought.
That’s a harder problem than anyone is selling at conferences right now. And it means the teams that think “adding agent support” is a single project are underestimating the scope by an order of magnitude.
The Window Is Closing
The companies waiting for a “winner” in the protocol wars will spend 2027-2028 rebuilding their entire commerce stack to catch up. The companies starting now get 18 months of integration experience before competitors begin.
This is the 1995 browser wars. Companies said “let’s wait and see if Netscape or IE wins.” The ones who picked early and built for the web became Amazon, eBay, and Google. The ones who waited aren’t around anymore.
Bain & Company estimates a $300-500 billion market by 2030. That’s 15-25% of total U.S. online retail up for grabs. The play isn’t picking a protocol. The play is building the infrastructure that makes any protocol a straightforward integration.
And the play nobody else is making? Understanding that agentic commerce isn’t a wholesale replacement of how people shop. It’s a new channel that serves a specific type of purchase behavior. Build for that nuance and you’ll be ahead of every team that treated this as a simple “add agent support” sprint.
The agents are here. They just can’t find you.
And by the time you notice the traffic you’re not getting, your competitors will have been training agents on their catalog for a year.